3D Handwriting Analysis
Crossover Resolution over Complex Images and Development of Corpus
Lead: Razdan, Anshuman
Sponsor: Miscellaneous Federal Government Agencies
Date: 07/26/2006 - 07/26/2007
Abstract
The objective of this work is to research and develop algorithms for detecting Arabic handwriting scribbled over complex documents consisting of typed text, schematics, drawings, etc. Scribbles in the margin area are of interest as well. The project seeks to develop a novel technique to synergistically fuse various page decomposition algorithms to lift handwritten scribbles as part of the document triage process.
 |
| Representation of the 2D bitmap of line scan (left). Representation of the conversion process from 2D bitmap to 3D voxel (volume) data (middle). Representation of the extraction of 3D surface (triangle mesh) from volumetric data (right). |
The 3D character recognition approach uses an understanding of direction of the written line, density, and other cues to extract information similar to the temporal information about the lines that compose characters provided by the real time online techniques. The 3D method separates the lines and loops that define the characters, converts them into chain codes, and creates a surface model that identifies the volume of the ink that defines a given character. This use of the derived chain codes and volume models provides valuable information that can be applied to the task of separating and correctly identifying the lines that describe a single unit or character. Though not sufficient for full OCR, this technique can provide valuable new information to assist paleographers in analyzing and identifying complex characters and separate overlapping lines during translation.
 |
| Overlapping components are shown in different colors (bottom). |
 |
| Detail of document scan bitmap and 3D surface model extracted from bitmap showing the line overlap that created the characters. |
